KD wrote: They scale very good, both in data and in constraints. In my experience, local search (tabu search, SA, ...) is the best up-scaling algorithm out there. Getting StartingSolutionInitializer right and setting a minimalAcceptedSelection so a step is done every 1-2 seconds is critical though. Join the webinar next week: http://blog.athico.com/2011/05/drools-planner-webinar-on-wednesday.html If I have time I 'll include some graphs to prove how it scales out. Or try it yourself in the examples: run the *Importer (classes with main()) and they 'll output you how big each dataset it. Then use the *Benchmarker to get a graph which the results for each dataset. KD wrote: Publishing benchmarks of some of the commercial solvers is illegal under their license terms. Some notes: - Micro benchmarks are generally worthless and cheatable (see JavaOne/Devoxx presentations of Cliff Click, Joshua Bloch, ...). - Benchmarking N-queens is stupid: it's not NP complete. It's very cheatable and besides the Drools Planner implementation is an unoptimized tutorial. - Benchmarking pure TSP is ok but short-sighted. You need to be able to scale in data AND constraints. (I am working on a TSP example in Planner but it's not finished yet). - Competitions such as ITC2007, RAS, ... are great benchmarks: they use real-world data, real constraints, ... I finished 4th in ITC2007 track 1 (much has improved since then) and all of the other finalists were researchers that wrote their experimental solvers from scratch as far as I know. -- View this message in context: http://drools.46999.n3.nabble.com/DROOLS-PLANNER-computation-times-and-co... Sent from the Drools: User forum mailing list archive at Nabble.com.